On Ollama, an uncensored model is a GGUF you download and run on your own card: llama2-uncensored, the Dolphin family, wizardlm-uncensored, the abliterated DeepSeek distills. Below is each of those tags as the library showed them on 27 October 2026, with its weights, beside the matching or newer build on this key, its live per-million price and the cost of 100 messages. The closing section is the two settings that move Open WebUI or AnythingLLM from Ollama to the key. To try a build before any setup, the chat gives three free turns a day with no account, and the Telegram bot does the same.
If the two terms are new to you, Abliterated vs uncensored sets out what separates the two in a single page.
What the Ollama library calls uncensored
Run the library’s "uncensored" search and, as of 27 October 2026, you get eight official tags plus about twelve from community namespaces. The table holds the official ones. The tag links to its page on ollama.com, and the size column repeats what that page prints. None of these tags has been pushed in the last year.
| Tag | Weights | From the Ollama page |
|---|---|---|
llama2-uncensored | 7B at 3.8 GB or 70B at 39 GB | Llama 2 tune from George Sung and Jarrad Hope; 2.7M pulls, 2 years since the last push |
dolphin-llama3 | 8B at 4.7 GB or 70B at 40 GB; 8K context | Eric Hartford’s Dolphin 2.9 over Llama 3; 2.2M pulls, 2 years since the last push |
dolphin-mistral | 7B at 4.1 GB; 32K context | Dolphin 2.8 over Mistral 7B; 1.8M pulls, 2 years since the last push |
dolphin3 | 8B at 4.9 GB; 128K context | Dolphin 3.0 over Llama 3.1 8B; 4.2M pulls, 1 year since the last push |
wizardlm-uncensored | 13B at 7.4 GB; 4K context | 666.5K pulls, 2 years since the last push |
huihui_ai/deepseek-r1-abliterated | seven sizes, 1.5B at 1.1 GB up to 70B at 43 GB; 128K context | The DeepSeek R1 distills with huihui_ai’s abliteration; 748.7K pulls, 1 year since the last push |
Also in the results, and absent from the table because nothing on this key stands in for them: everythinglm at 13B with a 16K context, wizard-vicuna-uncensored from 7B up to 30B, and dolphin-mixtral in its 8x7B and 8x22B forms. On the community side the results were mostly Qwen 3.8 and Gemma 4 abliterations under personal names, eriktonon/qwen3.8-uncensored at 27B and at 9B being one of them.
ollama run llama2-uncensored# or any tag in the table: ollama run huihui_ai/deepseek-r1-abliterated:14bEach tag matched to a hosted build
For every tag above there is an id on this key from the same family or a later generation, answered over the API and billed per token. The 70B llama2-uncensored download is 39 GB; llama-3.3-70b-abliterated gives you Llama 3.3 at that size, refusal direction cut out. Library Dolphins top out at 8B; dolphin-mistral-24b-venice runs the same recipe on Mistral Small 24B. In place of the R1 distills there is deepseek-v4.1-flash-abliterated: today’s DeepSeek, edited rather than retrained.
| Tag | Id on this key | Input / M | Output / M | 100 messages |
|---|---|---|---|---|
llama2-uncensored | llama-3.3-70b-abliterated | $2.80 | $2.80 | $0.64 |
dolphin-llama3 | llama-3.3-70b-abliterated | $2.80 | $2.80 | $0.64 |
dolphin-mistral | dolphin-mistral-24b-venice | $0.50 | $2.25 | $0.17 |
dolphin3 | dolphin-mistral-24b-venice | $0.50 | $2.25 | $0.17 |
wizardlm-uncensored | qwen3.8-27b-uncensored | $1.2125 | $3.6375 | $0.35 |
huihui_ai/deepseek-r1-abliterated | deepseek-v4.1-flash-abliterated | $0.8488 | $1.6975 | $0.22 |
The verdict column comes from the real-model check: once a day a fixed prompt goes to every id and the answer is compared with the model named on the label. Nothing like it exists for a local pull, since the file on your disk is simply the file you downloaded.
| Wire formats | |||||
|---|---|---|---|---|---|
| Llama 3.3 70B Abliterated | Abliterated | $2.80 | $2.80 | 33K | |
| Dolphin Mistral 24B Venice Edition | Uncensored | $0.50 | $2.25 | 33K | |
| Qwen3.8 27B Uncensored | Uncensored | $1.2125 | $3.6375 | 33K | |
| DeepSeek V4.1 Flash Abliterated | Abliterated | $0.8488 | $1.6975 | — |
| Id | Verdict today | Window |
|---|---|---|
llama-3.3-70b-abliterated | Steady since Oct 29 | 32,768 tokens |
dolphin-mistral-24b-venice | Steady since Oct 23 | 32,768 tokens |
qwen3.8-27b-uncensored | Steady since Oct 29 | 32,768 tokens |
deepseek-v4.1-flash-abliterated | Steady since Oct 19 | not published |
Family pages hold the rest: every uncensored Llama for the Llama ids, the DeepSeek list for the DeepSeek id, and the Qwen list for the Qwen id. No install is needed to try any of them: the same ids answer in the browser chat, three free turns a day without an account.
The abliterated DeepSeek on Ollama, and the one on this key
Searches for "ollama deepseek abliterated" land on huihui_ai/deepseek-r1-abliterated. On 27 October 2026 its page offered seven sizes between 1.5B and 70B, every one with a 128K window, 748.7K pulls and a push one year back. At 20 GB for the 32B quant and 43 GB for the 70B, a laptop realistically runs the 7B, 8B or 14B distill.
Here the DeepSeek ids are deepseek-v4.1-flash-abliterated, V4.1 Flash as it stands today minus its refusal direction, and the quicker lane for the identical build, deepseek-v4.1-flash-abliterated-fast. Both are DeepSeek proper, not distills. Ollama’s OpenAI-compatibility doc puts its own endpoint at http://localhost:11434/v1, so moving the request over means swapping the host and the model id and nothing else.
# Ollama's own OpenAI-compatible endpoint, from its docscurl http://localhost:11434/v1/chat/completions \ -H "Content-Type: application/json" \ -d '{"model": "dolphin-llama3:8b", "messages": [{"role": "user", "content": "Hello"}]}'# The same request against this key: change the host and the model idcurl https://api.unrestrictable.ai/v1/chat/completions \ -H "Authorization: Bearer sk-unr-live-…" \ -H "Content-Type: application/json" \ -d '{"model": "llama-3.3-70b-abliterated", "messages": [{"role": "user", "content": "Hello"}]}'The second call is charged per token at the rate in the table and asks for no download, no VRAM and no choice of quant. Keys begin with sk-unr-, come from the console and are displayed one time only.
Where the Dolphin and Llama 2 tags stand
Dolphin 2.9 on Llama 3 is what dolphin-llama3 ships, in 8B and 70B with an 8K window. llama2-uncensored goes back further, Llama 2 in 7B and 70B with a 4K window. Neither page has moved in two years. Both still load, and as a 7B on a laptop either does the job.
At 70B the hosted id is the sensible choice. The GGUF alone is 39 to 40 GB before anything runs. llama-3.3-70b-abliterated takes Llama 3.3 70B Instruct and applies the huihui-ai abliteration; its window is in the table above and its per-token rate is on the matching row. The last column there is what an evening of 100 messages comes to, with no power bill and no download.
If Dolphin is what you want, dolphin-mistral-24b-venice is Cognitive Computations’ Dolphin on Mistral Small 24B, more than three times the size of the 7B tag; see the table above for its window.
Open WebUI and AnythingLLM: swap Ollama for the key
Each app already reaches Ollama over an OpenAI-compatible connection, and each accepts another connection of the same type. Only the URL and the key differ. Once saved, the picker reads /v1/models and lists every uncensored id the key can see.
Open WebUI
Go to Settings > Admin > Connections and add an OpenAI API connection as the connection doc describes. The Ollama connection can stay enabled so both sets of models share one picker.
URL https://api.unrestrictable.ai/v1API Key sk-unr-live-…For a trimmed picker and a preset per build, the Open WebUI guide walks through it.
AnythingLLM
Open Settings and change the LLM provider from Ollama to the entry AnythingLLM’s docs name OpenAI (generic), the one meant for developers, then enter the base URL, the key and a model id:
Base URL https://api.unrestrictable.ai/v1API key sk-unr-live-…Model llama-3.3-70b-abliterated- Base URL
- https://api.unrestrictable.ai/v1Includes /v1; nothing after it
- Key
- sk-unr-live-…Goes in the Bearer header
- Model
- llama-3.3-70b-abliteratedAny id /v1/models returns
Test the key from a terminal before you touch either app. The request is free and prints the exact ids the picker is about to show.
curl https://api.unrestrictable.ai/v1/models -H "Authorization: Bearer sk-unr-live-…"The same URL and key work in Janitor AI, SillyTavern and the rest of the roleplay front-ends; the docs page names the fields for each. Developers who want the raw endpoint will find it there too.
Questions
The tag Ollama is serving is a stock instruct model. llama3, mistral, qwen and gemma from the library keep their alignment, and AnythingLLM simply forwards your prompt to whatever tag you picked. Either pull one of the uncensored tags in the first table and choose it, or point the provider at this API and choose an abliterated id. With the refusals edited out of the weights there is no system prompt to write. See AI with no restrictions.
On a laptop, the 7B and 8B tags fit, and dolphin3 has the longest window of them at 128K. A 70B means a 39 to 43 GB download; the hosted ids in the second table give you that size class per token with nothing to install. Every build on this key is ranked on the leaderboard. See the uncensored LLM leaderboard.
Only from the community. The search on 27 October 2026 turned up abliterated Qwen 3.8 tags like eriktonon/qwen3.8-uncensored in 9B and 27B, with no official one. This key serves qwen3.8-27b-uncensored as the general-purpose build, and the other Qwen tunes sit on the uncensored Qwen page. What is never allowed is spelled out on the acceptable use page. See the uncensored Qwen page.
It is. On 27 October 2026 the library still carried the tag, 2.7M pulls, 7B and 70B, last pushed two years earlier. Being Llama 2 it has a 4K window; the current Llama at that size on this key is llama-3.3-70b-abliterated.
It is the R1 distill set: Qwen and Llama models trained on R1 output, then abliterated by huihui_ai, in sizes from 1.5B to 70B. The full 671B DeepSeek does not fit a consumer GPU. deepseek-v4.1-flash-abliterated on this key is a true DeepSeek build rather than a distill, and the daily check records whether it keeps answering like one. See the daily real-model check.
It can. Open WebUI holds any number of connections. Leave the Ollama connection in place for your local tags and add https://api.unrestrictable.ai/v1 with your key as an OpenAI API connection; the picker then shows both lists.
Every model above is on the uncensored model list with its live price per million tokens. Per-token pricing sets the whole catalog side by side, and the family directory groups the builds by base model.